WeChat Mini Game Helper

Problem Solved

When AI assistants work on WeChat Mini Games, they often need MCP tools for tasks such as run_game, log inspection, and game-related workflows. The practical blocker is rarely code writing; it is the fragmented environment: Node.js may be missing or too old, the MCP package may not be installed, the mcp.json file may differ across AI hosts, or a configured entry may exist but not be loaded by the active assistant. When multiple agents and configuration directories coexist, troubleshooting becomes noticeably harder. This skill turns that setup path into a controlled, inspectable flow.

How It Works

It does not depend on a specific IDE plugin stack. It assumes Node.js 18 or newer and uses the @weadmin/weixin-minigame-helper-mcp service. The flow starts by trying the MCP call directly. If that fails, helper scripts check Node status, package resolution, candidate configuration files, and a recommendedAction. The decision evidence is concentrated in script output JSON fields, so an AI assistant can follow a branch without loading all script sources into context. When the host is confidently detected, the skill can install or configure the MCP entry. When the host is weak or ambiguous, it returns a manualSnippet so the assistant can give the user a JSON snippet to paste, rather than guessing the wrong config file.

Boundaries and Caveats

This is an MCP-only workflow and does not fall back to shell-based previewing when MCP is unavailable. A vscode signal is weak: it may be vanilla VSCode, a fork such as Cursor, Windsurf, or CodeBuddy, or an internal variant. When targetHostCaveat or ask-user appears, manual pasting is the safer path. If the MCP package can start but the active assistant still cannot call it, the issue usually points to a mismatch in config path, transport mode, or host variant. After configuration, verify the MCP server and call the tool directly instead of recommending an editor restart as a normal step.